Instantiate constraint entries when updated to equal bounds

[EugeneFlesselle]

Checking isSub on the resulting bounds may have introduced new constraints, which might allow us to replace the type parameter entirely.

Adapted from #20200
@jchyb can you test if this also addresses the performance issue in #20120 ?

[EugeneFlesselle]

These changes also solve #19955 independently of the changes in #20175, with I think a much more general solution.

[jchyb]

Sure, let me check

[jchyb]

It doesn't seem to fix the performance here, unfortunately (typer still takes vastly more time in T1, which is the val that summons the implicit with the intersection type which causes the TypeBounds constraint)

[EugeneFlesselle]

I couldn't figure out how to test the changes with -Yprofile-trace.
Async profiler reports a 59% improvement in the Typer phase though.

@jchyb could you run it again ?

[odersky]

LGTM!

[jchyb]

There is a bit of an improvement:

(T1 takes 130ms)
where before this change T1 took 200ms:

but I still get the best results when putting case TypeBounds(lo, hi) if lo eq hi => lo in instType (which might be incorrect of course, but serves as a nice point of comparison):

(the screenshots aren't to scale, T1 here takes 40ms - but it's very possible I may be doing something wrong here of course)

[EugeneFlesselle]

Perfomance measurements:

Before changes:

• T0: 52ms
• T1, T1bis: 189, 79 ms
• T2, T2bis: 5, 4 ms

After changes:

• T0: 28ms
• T1, T1bis: 13, 25 ms
• T2, T2bis: 9, 6 ms